FormalPara Definition

Network effects arise when the value a customer derives from a good or service grows as other customers adopt compatible products.

Network effects arise when the value a customer derives from a good or service grows as other customers adopt compatible products (Katz and Shapiro 1985). There are two types of network effects: direct and cross-group. Direct network effects emerge when each customer’s utility increases with the number of other customers who use the same product or technology. Here, utility is gross of the price paid (if any). This is meant to rule out the ‘pecuniary effects’ criticized by Liebowitz and Margolis (1994), that is, effects operating solely through price (more participants leading to lower prices, hence benefiting existing participants). Examples of direct network effects include fax machines; instant messaging services such as AOL, MSN, Yahoo!, or QQ in China; and social networks such as Facebook and LinkedIn.

Cross-group network effects occur when there are at least two different customer groups that are interdependent, and the utility of at least one group grows as the other group(s) grow. The most obvious examples are game consoles, which become more valuable to consumers with the emergence of gaming applications; or operating systems such as Windows or Android, which grow in utility with the growth of applications. Conversely, the utility derived by developers of games or operating system applications grows with the growth of end-users. Cross-group network effects also exist between buyers and sellers on eBay; content providers and end-users in VCRs and high-definition DVD formats such as Blu-Ray; merchants and consumers for payment systems such as Visa, American Express and PayPal.

The academic literature on network effects was pioneered in economics by David (1985), Farrell and Saloner (1985, 1986), Katz and Shapiro (1985, 1986) and Arthur (1989). Most of the early literature focused on direct network effects and used the term ‘network externalities’. Network effects were also sometimes known as demand-side economies of scale. They are the counterpart of supply-side economies of scale, which occur when the average cost of producing an additional unit is decreasing in relation to the total number of units produced, or when the quality of an additional unit is increasing in relation to the number of units produced. Many ‘traditional’ industries exhibit supply-side economies of scale (e.g., car and steel manufacturing), but few of them have network effects. In contrast, network effects are pervasive in ‘new economy’ industries, particularly information and communication technologies. Some of them exhibit both supply-side economies of scale and network effects (e.g., operating systems such as Google’s Android and Microsoft’s Windows).

While economies of scale are inherently bounded, network effects can exhibit increasing returns. The average production costs of most physical products decrease until capacity constraints require the building of a new plant, which leads to a sharp increase in average costs. For digital products (e.g., Windows), they can go to zero, but no lower. In turn, while any user’s willingness to pay for Windows eventually stops increasing (after 10 or 20 or 100 software applications), the profits of third-party application developers can increase without bound with the number of Windows users.

Market Tipping

For strategic management, the importance of network effects is the possibility that markets can ‘tip’, that is, lead to the outcome in which only one technology, product or service emerges as the clearly dominant player while others are marginalized or even disappear (Shapiro and Varian 1998). Indeed, network effects generally imply that there is value for users to coordinate on adopting the same technology or platform. But not all markets with network effects will tip. Three additional conditions are necessary for tipping to occur: (1) the value of the network effects must outweigh the benefits of differentiation for users; (2) users must have high ‘multi-homing’ costs (i.e., the costs of adopting two or more technologies – see Crémer et al. 2000); and (3) users must have high switching costs (i.e., the costs of abandoning one technology in favour of another). If any of these three conditions fail, multiple technologies with significant market shares may coexist. For example, despite strong direct network effects, the instant messaging market in the United States remained an oligopoly (MSN, AOL, Yahoo! all have market shares larger than 20%) for 15 years, mainly because it was very easy for consumers to use multiple instant messaging accounts simultaneously. In contrast, the PC operating system market tipped to Windows in the early 1990s, in part because of the high switching costs to move between alternative platforms (PCs and Macs), and because of the multihoming costs for application developers, which led to a high barrier to entry, favouring Windows.

Finally, when markets with network effects do tip, this is not a guarantee of permanent success. Both external and internal shocks can cause markets with strong network effects to atrophy (Cantillon and Yin 2010). New technologies (mainframe computers versus PCs and commodity servers), substitution (fixed price sales and internet search versus online auctions), product failures (Internet Explorer security holes versus Firefox browsers), and raising prices above customers’ willingness to pay (Netscape browsers versus Microsoft browsers) can reintroduce competition in markets which appeared to have tipped.

Multiple Equilibria and Expectations

One of the most important consequences of network effects (both direct and cross-group) is the possibility of multiple equilibrium market outcomes (keeping all else equal). Which equilibrium will arise is crucially determined by expectations of the market participants involved. For instance, if each individual participant expects one product or technology to win (i.e., to draw the largest number of users), then each participant has a strong incentive to adopt that product or technology only, leading to self-fulfilling expectations.

Consequently, markets with network effects can exhibit inefficient outcomes; that is, the best technology or product (in the sense of total social value created) may not win. Instead, winners may be determined by historical accident (cf. the QWERTY keyboard example studied by Arthur 1989) or path dependence (i.e., the first technologies to market have an advantage). Furthermore, tipping may occur as the result of sudden changes in market participants’ expectations (e.g., holding expectations biased in favour of one technology over another) and without any change in the underlying market ‘fundamentals’. For example, in a classic battle over network effects in the internet browser business, webmasters were unwilling to commit to Netscape’s browser (and build their websites around Netscape’s technology), despite an early mover advantage and a dominant market share. Once Microsoft signalled an unswerving commitment to winning in browsers, it successfully froze webmasters until the outcome of the battle was clear (Cusumano and Yoffie 1998).

Indirect Network Effects and Multi-Sided Platforms

The distinction between direct and cross-group (sometimes called ‘indirect’) network effects was made early on (Farrell and Saloner 1985; Church and Gandal 1992), but the systematic study of firms serving multiple and interdependent customer groups did not develop until the early 2000s. In the following decade, there was an explosion of work on cross-group network effects, ‘multi-sided platforms’ and the implications for strategic management. Notable early contributions include Schmalensee (2002), Caillaud and Jullien (2003), Evans (2003), Rochet and Tirole (2003, 2006), Parker and Van Alstyne (2005), Armstrong (2006), Evans et al. (2006) and Hagiu (2006).

Much of the literature on multi-sided platforms (MSPs) focused on pricing; that is, the choice by MSPs to subsidize the participation of one or more sides and make most of their profits by charging other sides. An obvious example was eBay’s decision to charge sellers, but not buyers. One of the core pricing principles that emerged was that MSPs should charge lower fees and derive lower profits from the side(s) that generate(s) relatively stronger cross-group network effects for the other side(s) (Parker and Van Alstyne 2005; Armstrong 2006; Hagiu 2006; Rochet and Tirole 2006). More recent research has explored non-price, economic and strategic issues, such as how firms should design a platform to drive stronger cross-group network effects (e.g., Parker and Van Alstyne 2008; Hagiu and Jullien 2011), and how to set rules for governing a platform (e.g., Boudreau and Hagiu 2009).

Subtleties of Network Effects

While the definition of network effects is straightforward, their nature can be quite subtle in practice. First, the difference between network effects and economies of scale is not always clear cut. Consider the example of search engines like Google and Bing. Clearly, they exhibit cross-group network effects: the advertisers’ willingness to bid for sponsored search keywords has increased with the size of user traffic. But do search engines also exhibit direct network effects? Indeed, the quality of the service provided to users (accuracy of search results) has improved with the number of users. With several hundred billion searches per year, Google could improve its search algorithms more quickly than competitors. But this was an instance of economies of learning and scale (Varian 2008). Still, users who understand this mechanism may start making their search engine choice decisions (e.g., Google versus Bing) based on the number of other users they think are using each search engine. In fact, just as with many information goods, the quality of search can be hard to ascertain directly. In this case, economies of learning and scale can be transformed into network effects by the user decision-making process. Of course, even when such direct network effects appear in search, end-user switching costs remain quite low, so tipping is not guaranteed.

Second, network effects are not always exogenously given: their existence and magnitude may instead be endogenously determined by firms’ strategic choices.

Whether network effects are present or not depends on the relationship between the focal firm and other market participants, particularly in the context of multi-sided platforms (Hagiu and Wright 2011). For example, Microsoft Windows exhibits cross-group network effects because (a) it enables direct interactions between users and application developers and (b) both users and developers are affiliated with Windows by making relationship-specific investments in it (users purchase and learn how to use Windows; developers make their apps using Windows application programming interfaces, or APIs). In contrast, Dell’s or Lenovo’s PC hardware does not exhibit cross-group network effects. Although they are at least as important as Windows in enabling users–developer interactions, application developers do not make any investments specific to Dell’s and Lenovo’s hardware.

The magnitude of cross-group network effects depends on the contracts implemented by multisided platforms, in particular on the extent to which they enable direct interactions (Hagiu and Wright 2011). For instance, Amazon functions as an online retailer for some of its products and as a marketplace for others. The online retailer model exhibits weaker cross-group network effects: Amazon buys products from suppliers and resells them to end-users in its own name. These suppliers care about the number of consumers shopping at Amazon only insofar as Amazon does not take full inventory risk for their products. In contrast, in the marketplace model, Amazon enables third-party suppliers to sell directly to consumers, using Amazon’s website and shipping capabilities. In this latter model, the value for third-party sellers much more clearly and directly increases in accordance with the number of Amazon customers – and vice versa. Another example is online dating. Match.com allows anyone to join its two-sided platform, while eHarmony screens its participants carefully. Both exhibit cross-group network effects between men and women, but, at equal numbers, eHarmony arguably generates stronger network effects because of the higher ‘quality’ (i.e., suitability for long-term relationship) of its members (Halaburda and Piskorski 2010).

Network effects have been a cyclical academic industry: interest in the phenomena peaked in the late 1990s, with the dot-com boom. Early in the next decade, network effects were often discounted until a new literature emerged on MSPs and cross-group network effects. Looking forward, it is the potential to drive network effects endogenously through strategy that should place them at the core of the strategic management literature.

See Also